Date of Award


Document Type


Degree Name

Doctor of Philosophy (PhD)

First Advisor

Dr. Abdollah Homaifar


Sensors play a critical role in the development of intelligent systems. Intelligent agents are equipped with an array of sensors to acquire information about themselves and the environment to make a reasonable decision. Information extracted from the sensor data is often characterized by uncertainty. Modeling and reasoning under such uncertainty poses a great challenge. Multisensor data fusion is a viable approach to address this problem. The Dempster Shafer (DS) theory of belief functions, also known as the evidence theory, is a well-known data fusion formalism due to its close relationship with other mathematical theories of uncertainty and its elegant way for the representation and combination of uncertain information. In this dissertation, the theory of belief functions is proposed to deal with uncertainty in sensory data with application in target classification. One of the mainstays of the DS theory is the rule of combination. The DS rule of combination can be used to aggregate uncertain pieces of information (represented by belief functions). However, the DS combination rule suffers a major setback of counter-intuitive results when it is required to pool belief functions that are highly conflicting with one another. An open problem is how to manage high conflict among the various pieces of information effectively. One way to confront this problem is to assign weighting factors to the participating pieces of information to control their contribution to the combined belief function. In the absence of the ground-truth data, the allocation of a weighting factor is a challenging problem. In this dissertation, we proposed a weighting factor based on the consensus belief function. In this approach, the relative weights of the participating belief functions are positively correlated to their closeness with the consensus belief function. This will ensure that the impact of the conflicting pieces of information on the fusion results is reduced accordingly. Furthermore, we formulated a target classification problem in which every attribute of the observed target induces a belief function. With the availability of information regarding the various sources of belief functions, a reliability factor is estimated. Consequently, we proposed a reliability-credibility based Dempster Shafer rule of combination (RCDSRC) for the fusion 2 of the different uncertain pieces of information. This strategy involves a bi-criteria evaluation using a combination of the credibility degree and the reliability degree as the weighting factor before the deployment of the original DS rule of combination. In the evidence theory, a large number of information sources has been identified as one of the principal culprits of conflict and high processing cost. In this dissertation, we equally proposed the average pairwise discordance index (APDI) to reduce the number of information sources. Several real-world and synthetically generated problems were used to demonstrate the effectiveness of the proposed methods. We wrap up the dissertation with conclusions and future research directions.